Erosion Prediction Gaussian Process Regression Algorithm for Alumina and Chromia Reinforced Nickel-Based High-Velocity Oxy-Fuel Coatings

Abstract

Background:: Machine learning (ML) methodologies have demonstrated efficacy in the determination of erosion rates and material removal. In this context, a novel Erosion Prediction Gaussian Process Regression Algorithm (EPGPRA) was developed to predict the volumetric erosion in thermal spray coatings. Aim:: In this patent, a novel EPGPRA based model was developed to predict the volumetric loss of 30Al2O3 and 20Cr2O3 reinforced Ni-based coatings deposited using a high-velocity oxy-fuel (HVOF) process. Objective:: The objective of this patent is to develop a GPR model for the prediction of Ni-30Al2O3 and Ni-20Cr2O3 coatings. Methods:: Spraying powders were applied to the SS316L steel substrate in order to develop coatings. An erosion tester was used in order to investigate the wear resistance of HVOF-coated steel. The gathered experimental dataset is put to use in the construction of a powerful GPR model. The outcomes from GPR model were then measured against the values obtained from the experiments. To demonstrate the accuracy of the GPR model, the produced model is evaluated against various cutting-edge machine learning methods. Results:: This innovation was successful in terms of developing a new GPR model for wear prediction. The discrepancy between the actual and expected values is the smallest for Matern 5/2 (M5/2) GPR in the validation set. It was also lesser as compared to Ensemble Boosted Trees, Support Vector Machine, Linear regression, and Fine Tree. In terms of MSE, MAE, RMSE, and R2 the accuracy performance of the M5/2 GPR model was determined to be 9.8565×10-5, 0.0048884, 0.009928, and 0.93 correspondingly. Ni-Chromia coating performed better than the Ni-Alumina. Conclusion:: As per this patent, a novel EPGPRA-based model was developed, which is the better machine learning technique for wear prediction of Ni-based HVOF coatings.